Disentangle Estimation of Causal Effects from Cross-Silo Data

The article introduces an innovative architecture to estimate causal effects from data spread across different silos. The architecture’s design allows for the seamless transmission of model parameters enriched with causal mechanisms. This is achieved through a combination of shared and private branches, and the introduction of global constraints to mitigate bias in the missing domains. The proposed method has been tested on new semi-synthetic datasets, showing improved performance over existing methods.

Publication date: 5 Jan 2024
Project Page: Not provided
Paper: https://arxiv.org/pdf/2401.02154